Can we stop classifying database variables?

[markcmiller86]

So, I've raised this question before (before we had GitHub discussions 😉). When we process various types of variables from databases, we classify them...

visit/src/avt/DBAtts/MetaData/avtTypes.h

Lines 17 to 27 in 15e3799

	enum avtVarType
	{
	AVT_MESH = 0,
	AVT_SCALAR_VAR, /* 1 */
	AVT_VECTOR_VAR, /* 2 */
	AVT_TENSOR_VAR, /* 3 */
	AVT_SYMMETRIC_TENSOR_VAR, /* 4 */
	AVT_ARRAY_VAR, /* 5 */
	AVT_LABEL_VAR, /* 6 */
	AVT_MATERIAL, /* 7 */
	AVT_MATSPECIES, /* 8 */

This leads to code blocks of conditional logic for each type all over the place (in database readers, in expression processing, in metadata, in GUI, in CLI metadata, in CLI expressions...its prolific how these classification permeates our coding and interfaces) where we do something special for each type but where that specialness doesn't always seem worthwhile.

What if we just decided NOT to classify database variables into these types and instead each variable had some properties that we care about...

• int: number of components
• int: tensor rank
• bool: satisfies requirements for a coordinate field
• bool: continuous in range space (means the field is continuously varying in "value" or not (e.g. is discrete instead))
• templated type: range-space data type (e.g. int, double, etc.)
• int[7]/int[7]: units (maybe...we don't really do this now except as arbitrary strings...see SAF data model.)

• A Mesh plot of any (non-mesh) variable means to plot the mesh the variable is defined on.
• A scalar plot (e.g. Pseudocolor, Contour, Volume...) of any (non-scalar) variable means to plot its magnitude
• A Subset plot of any (non-subsetting) variable means to plot a contour plot of the variable's scalar interpretation
  • Subset works as FilledBoundary...no more Boundary plot either...just a swizzle on subet plot
• A Curve lot of any (non-curve) variable plots its histogram (of its scalar interpretation)
  etc., etc.

All plots would plot all variables. I think we can do reasonable things when users attempt to combine a variable an a plot in a way that doesn't necessarily make the most sense. There's no trimming lists of variables based on a classification...though I could maybe see some argument for using tensor rank where plots indicate which tensor ranks they support and variables are matched to them (actually, I think that is a problem with our current design as well, the knowhow about which classes of variables a plot supports is not part of plots...its scattered about in all the conditional logic that processes lists of variables). And, do our users really not understand their data so much that they could select a combination of variable and plot and then not understand the result they are presented with?